Post on 18-Feb-2018
Network Biology- part IV
Jun Zhu, Ph. D.
Professor of Genomics and Genetic Sciences
Icahn Institute of Genomics and Multi-scale Biology
The Tisch Cancer Institute
Icahn Medical School at Mount Sinai
New York, NY
@IcahnInstitute
http://research.mssm.edu/integrative-network-biology/
Email: jun.zhu@mssm.edu
Association
networks
Probabilistic causal
networks
Biological details revealed
Data required to train models
Biological networks/pathways
1. How do genes in the same
module interact?
2. How do genes in different
modules interact?
3. Can we make causal
inferences to elucidate
signaling pathway for
disease targets?
4267 top genes in BxH liver female rescan qtl overlap (num(p(GGC)<1e-15)>100 ~abs(cor)>0.5886)
What are Bayesian networks? Association vs Causality
From Stephen Friend
A simple biological question: are there
causal/reactive relationships?
A Bayesian network approach:
Best model
What are Bayesian networks?
• A Bayesian network is an expert system that captures
all existing knowledge;
• They are also called belief networks, Bayesian belief
networks, causal probabilistic networks;
• A Bayesian network consists of
• a directed acyclic graph (a set of nodes and directed edges
connecting nodes)--DAG
• A set of conditional probability tables (for discrete data) or
probability density functions (for continuous data)
A Bayesian network
C
A B
F
D
( | , )p C A B
(D | )p B
E (E | )p B
(F | C)p
DAG Conditional
probability tables
Bayesian network
• A tree is a Bayesian network
C
A
B
F
D E
Bayesian network
C
A B
F
D E
• A Bayesian network is not a tree
Bayesian network
• Conventional Notations
( ) ( | ( ))i i
i
p p A pa AA
1 2 n{A ,A , ,A } are nodes.A
( ) is the joint probability of nodes .p A A
( ) are parent nodesof .i ipa A A
Bayesian network
C B
A
D E
• A diverging structure
out-degree =4
Bayesian network
D
A B C
• A converging structure
in-degree =3
Bayesian network
• Why a DAG is required?
( ) ( | ( ))i i
i
p p A pa AA
• It is guaranteed that there is a node Aj in a DAG that
has no child.
j
j
j
( ) ( \ { }) ( | \ { })
( \ { }) ( | ( ))
( ( | ( )))* ( | ( ))
j j
j j
i i j
i j
p p A p A A
p A p A pa A
p A pa A p A pa A
A A A
A
Bayesian network: usages
• Bayesian networks can be used to predict outcomes
or diagnose causal effects (if structures are known)
• Bayesian networks can be used to discover causal
relationships (if structures are not known)
Bayesian network: an example
Alarm
burglar Earthquake
Phonecall
Radio Internet
• A burglar alarm system
Bayesian network: a classifier
• What is a naïve Bayes net
C B
A
D E
( (A)
(( | , , )
(
p A,B,C,D,E)= p(B | A)p(C | A)p(D | A)p(E | A)p
p A,B,C,D,E)p A B C D
p B,C,D,E)
Bayesian network
B
A
C
• How to train a Bayesian network
A=a1 A=a2 A=a3
B=b1 7 12 25
B=b2 20 30 28
B=b3 25 20 6
A=a1 A=a2 A=a3
C=c1 15 8 20
C=c2 11 25 18
C=c3 27 10 16
Bayesian network
B
A
C
• How to construct a Bayesian network? Enumerating
possible structures
B
A
C B
A
C B
A
C
B
A
C B
A
C B
A
C
Bayesian network
• How to construct a Bayesian network? Enumerating all
possible structures is impossible
,NN N is thenumberof nodes
Bayesian network
• How to construct a Bayesian network? Heuristic approach
xi
Pa1 Pa2 Pan
xi
Pa1 Pa2 Pan
xi
Pa1 Pa2 Pan
xi
Pa1 Pa2 Pan
X Pan+1 Paj
a b c
Bayesian network
• How to construct a Bayesian network? Heuristic approach
D
A B
Parameters to estimate=3x3x3
D
A B C
Parameters to estimate=3x3x3x3
Bayesian network
• How to construct a Bayesian network? Heuristic approach
( | ) ( )( | )
( )
p D M p Mp M D
p D
ˆBIC 2ln ( ) ln( )
: number of samples
: number of parameters toestimate
p D | M k n
n
k
Bayesian network
• How to construct a Bayesian network? averaging
Zhu et al., PLoS CompBio, 2007
Zhu et al., Nature Genetics, 2008
Bayesian network • How to construct a Bayesian network? Enforcing DAG
after averaging
1. Calculate shortest distance
2. Identify loops
3. Remove the weakest link in a loop
4. Go to step 1
Zhu et al., PLoS CompBio, 2007
Bayesian network
• How to construct a Bayesian network? Upper limit on
in-degree
Parameters to estimate=
xi
Pa1 Pa2 Pan
13n
Bayesian network
• Continuous vs discrete models
• Discrete model is faster, easier to capture high order
interactions
• Any discretization lost information
Bayesian network
• Missing information
B
A
C B
X
C
A
B
X
C
A
Bayesian network
• Biological network is context specific
• Bayesian network is just a snapshot under a specific
condition
Other ways to infer causal networks
• Boolean network, Graphic Gaussian model
• Conditional Mutual Information
• ODE model
• Structural equation
modeling by differential equations
uxfdtdx )(/
Response function
Observed states
Perturbation
Gardner et al, Science, 2003
modeling by ordinary differential
equations (ODE)
▶ Assume static state dx/dt=0
▶ Assume linear relationships f(x)=Ax
0uAx
regulatory matrix
response matrix
01 uAx
ODE: advantages and disadvantages
▶ Advantages:
• Simple
• Can model feedback loops
▶ Disadvantages:
• Need large amount of data
• Need even more data to capture non-linear
relationships
Aknowledgements Mount Sinai
Genomics Institute
Eric Schadt
Bin Zhang
Zhidong Tu
Charles Powell
Patrizia Casaccia
Zhu lab
Seungyeul Yoo
Eunjee Lee
Li Wang
Luan Lin
Quan Long
•Icahn Institute of Genomics and Multiscale Biology,
Icahn School of Medicine at Mount Sinai
•Janssen
•Canary Foundation
•Prostate Cancer Foundation
•NIH
•NCI
Supported by:
Boston University
Avrum Spira
Joshua Campbell
U Washington
Roger Baumgarner
Berkerley
Rachel Brem
Princeton
Lenoid Kruglyak